1
# Copyright (C) 2007 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
"""Indexing facilities."""
23
'GraphIndexPrefixAdapter',
27
from bisect import bisect_right
28
from cStringIO import StringIO
32
from bzrlib.lazy_import import lazy_import
33
lazy_import(globals(), """
34
from bzrlib import trace
35
from bzrlib.bisect_multi import bisect_multi_bytes
36
from bzrlib.revision import NULL_REVISION
37
from bzrlib.trace import mutter
45
_HEADER_READV = (0, 200)
46
_OPTION_KEY_ELEMENTS = "key_elements="
48
_OPTION_NODE_REFS = "node_ref_lists="
49
_SIGNATURE = "Bazaar Graph Index 1\n"
52
_whitespace_re = re.compile('[\t\n\x0b\x0c\r\x00 ]')
53
_newline_null_re = re.compile('[\n\0]')
56
class GraphIndexBuilder(object):
57
"""A builder that can build a GraphIndex.
59
The resulting graph has the structure:
61
_SIGNATURE OPTIONS NODES NEWLINE
62
_SIGNATURE := 'Bazaar Graph Index 1' NEWLINE
63
OPTIONS := 'node_ref_lists=' DIGITS NEWLINE
65
NODE := KEY NULL ABSENT? NULL REFERENCES NULL VALUE NEWLINE
66
KEY := Not-whitespace-utf8
68
REFERENCES := REFERENCE_LIST (TAB REFERENCE_LIST){node_ref_lists - 1}
69
REFERENCE_LIST := (REFERENCE (CR REFERENCE)*)?
70
REFERENCE := DIGITS ; digits is the byte offset in the index of the
72
VALUE := no-newline-no-null-bytes
75
def __init__(self, reference_lists=0, key_elements=1):
76
"""Create a GraphIndex builder.
78
:param reference_lists: The number of node references lists for each
80
:param key_elements: The number of bytestrings in each key.
82
self.reference_lists = reference_lists
84
# A dict of {key: (absent, ref_lists, value)}
86
self._nodes_by_key = None
87
self._key_length = key_elements
88
self._optimize_for_size = False
90
def _check_key(self, key):
91
"""Raise BadIndexKey if key is not a valid key for this index."""
92
if type(key) != tuple:
93
raise errors.BadIndexKey(key)
94
if self._key_length != len(key):
95
raise errors.BadIndexKey(key)
97
if not element or _whitespace_re.search(element) is not None:
98
raise errors.BadIndexKey(element)
100
def _get_nodes_by_key(self):
101
if self._nodes_by_key is None:
103
if self.reference_lists:
104
for key, (absent, references, value) in self._nodes.iteritems():
107
key_dict = nodes_by_key
108
for subkey in key[:-1]:
109
key_dict = key_dict.setdefault(subkey, {})
110
key_dict[key[-1]] = key, value, references
112
for key, (absent, references, value) in self._nodes.iteritems():
115
key_dict = nodes_by_key
116
for subkey in key[:-1]:
117
key_dict = key_dict.setdefault(subkey, {})
118
key_dict[key[-1]] = key, value
119
self._nodes_by_key = nodes_by_key
120
return self._nodes_by_key
122
def _update_nodes_by_key(self, key, value, node_refs):
123
"""Update the _nodes_by_key dict with a new key.
125
For a key of (foo, bar, baz) create
126
_nodes_by_key[foo][bar][baz] = key_value
128
if self._nodes_by_key is None:
130
key_dict = self._nodes_by_key
131
if self.reference_lists:
132
key_value = key, value, node_refs
134
key_value = key, value
135
for subkey in key[:-1]:
136
key_dict = key_dict.setdefault(subkey, {})
137
key_dict[key[-1]] = key_value
139
def _check_key_ref_value(self, key, references, value):
140
"""Check that 'key' and 'references' are all valid.
142
:param key: A key tuple. Must conform to the key interface (be a tuple,
143
be of the right length, not have any whitespace or nulls in any key
145
:param references: An iterable of reference lists. Something like
146
[[(ref, key)], [(ref, key), (other, key)]]
147
:param value: The value associate with this key. Must not contain
148
newlines or null characters.
149
:return: (node_refs, absent_references)
150
node_refs basically a packed form of 'references' where all
152
absent_references reference keys that are not in self._nodes.
153
This may contain duplicates if the same key is
154
referenced in multiple lists.
157
if _newline_null_re.search(value) is not None:
158
raise errors.BadIndexValue(value)
159
if len(references) != self.reference_lists:
160
raise errors.BadIndexValue(references)
162
absent_references = []
163
for reference_list in references:
164
for reference in reference_list:
165
# If reference *is* in self._nodes, then we know it has already
167
if reference not in self._nodes:
168
self._check_key(reference)
169
absent_references.append(reference)
170
node_refs.append(tuple(reference_list))
171
return tuple(node_refs), absent_references
173
def add_node(self, key, value, references=()):
174
"""Add a node to the index.
176
:param key: The key. keys are non-empty tuples containing
177
as many whitespace-free utf8 bytestrings as the key length
178
defined for this index.
179
:param references: An iterable of iterables of keys. Each is a
180
reference to another key.
181
:param value: The value to associate with the key. It may be any
182
bytes as long as it does not contain \0 or \n.
185
absent_references) = self._check_key_ref_value(key, references, value)
186
if key in self._nodes and self._nodes[key][0] != 'a':
187
raise errors.BadIndexDuplicateKey(key, self)
188
for reference in absent_references:
189
# There may be duplicates, but I don't think it is worth worrying
191
self._nodes[reference] = ('a', (), '')
192
self._nodes[key] = ('', node_refs, value)
194
if self._nodes_by_key is not None and self._key_length > 1:
195
self._update_nodes_by_key(key, value, node_refs)
199
lines.append(_OPTION_NODE_REFS + str(self.reference_lists) + '\n')
200
lines.append(_OPTION_KEY_ELEMENTS + str(self._key_length) + '\n')
201
lines.append(_OPTION_LEN + str(len(self._keys)) + '\n')
202
prefix_length = sum(len(x) for x in lines)
203
# references are byte offsets. To avoid having to do nasty
204
# polynomial work to resolve offsets (references to later in the
205
# file cannot be determined until all the inbetween references have
206
# been calculated too) we pad the offsets with 0's to make them be
207
# of consistent length. Using binary offsets would break the trivial
209
# to calculate the width of zero's needed we do three passes:
210
# one to gather all the non-reference data and the number of references.
211
# one to pad all the data with reference-length and determine entry
215
# forward sorted by key. In future we may consider topological sorting,
216
# at the cost of table scans for direct lookup, or a second index for
218
nodes = sorted(self._nodes.items())
219
# if we do not prepass, we don't know how long it will be up front.
220
expected_bytes = None
221
# we only need to pre-pass if we have reference lists at all.
222
if self.reference_lists:
224
non_ref_bytes = prefix_length
226
# TODO use simple multiplication for the constants in this loop.
227
for key, (absent, references, value) in nodes:
228
# record the offset known *so far* for this key:
229
# the non reference bytes to date, and the total references to
230
# date - saves reaccumulating on the second pass
231
key_offset_info.append((key, non_ref_bytes, total_references))
232
# key is literal, value is literal, there are 3 null's, 1 NL
233
# key is variable length tuple, \x00 between elements
234
non_ref_bytes += sum(len(element) for element in key)
235
if self._key_length > 1:
236
non_ref_bytes += self._key_length - 1
237
# value is literal bytes, there are 3 null's, 1 NL.
238
non_ref_bytes += len(value) + 3 + 1
239
# one byte for absent if set.
242
elif self.reference_lists:
243
# (ref_lists -1) tabs
244
non_ref_bytes += self.reference_lists - 1
245
# (ref-1 cr's per ref_list)
246
for ref_list in references:
247
# how many references across the whole file?
248
total_references += len(ref_list)
249
# accrue reference separators
251
non_ref_bytes += len(ref_list) - 1
252
# how many digits are needed to represent the total byte count?
254
possible_total_bytes = non_ref_bytes + total_references*digits
255
while 10 ** digits < possible_total_bytes:
257
possible_total_bytes = non_ref_bytes + total_references*digits
258
expected_bytes = possible_total_bytes + 1 # terminating newline
259
# resolve key addresses.
261
for key, non_ref_bytes, total_references in key_offset_info:
262
key_addresses[key] = non_ref_bytes + total_references*digits
264
format_string = '%%0%sd' % digits
265
for key, (absent, references, value) in nodes:
266
flattened_references = []
267
for ref_list in references:
269
for reference in ref_list:
270
ref_addresses.append(format_string % key_addresses[reference])
271
flattened_references.append('\r'.join(ref_addresses))
272
string_key = '\x00'.join(key)
273
lines.append("%s\x00%s\x00%s\x00%s\n" % (string_key, absent,
274
'\t'.join(flattened_references), value))
276
result = StringIO(''.join(lines))
277
if expected_bytes and len(result.getvalue()) != expected_bytes:
278
raise errors.BzrError('Failed index creation. Internal error:'
279
' mismatched output length and expected length: %d %d' %
280
(len(result.getvalue()), expected_bytes))
283
def set_optimize(self, for_size=True):
284
"""Change how the builder tries to optimize the result.
286
:param for_size: Tell the builder to try and make the index as small as
290
# GraphIndexBuilder itself doesn't pay attention to the flag yet, but
292
self._optimize_for_size = for_size
295
class GraphIndex(object):
296
"""An index for data with embedded graphs.
298
The index maps keys to a list of key reference lists, and a value.
299
Each node has the same number of key reference lists. Each key reference
300
list can be empty or an arbitrary length. The value is an opaque NULL
301
terminated string without any newlines. The storage of the index is
302
hidden in the interface: keys and key references are always tuples of
303
bytestrings, never the internal representation (e.g. dictionary offsets).
305
It is presumed that the index will not be mutated - it is static data.
307
Successive iter_all_entries calls will read the entire index each time.
308
Additionally, iter_entries calls will read the index linearly until the
309
desired keys are found. XXX: This must be fixed before the index is
310
suitable for production use. :XXX
313
def __init__(self, transport, name, size):
314
"""Open an index called name on transport.
316
:param transport: A bzrlib.transport.Transport.
317
:param name: A path to provide to transport API calls.
318
:param size: The size of the index in bytes. This is used for bisection
319
logic to perform partial index reads. While the size could be
320
obtained by statting the file this introduced an additional round
321
trip as well as requiring stat'able transports, both of which are
322
avoided by having it supplied. If size is None, then bisection
323
support will be disabled and accessing the index will just stream
326
self._transport = transport
328
# Becomes a dict of key:(value, reference-list-byte-locations) used by
329
# the bisection interface to store parsed but not resolved keys.
330
self._bisect_nodes = None
331
# Becomes a dict of key:(value, reference-list-keys) which are ready to
332
# be returned directly to callers.
334
# a sorted list of slice-addresses for the parsed bytes of the file.
335
# e.g. (0,1) would mean that byte 0 is parsed.
336
self._parsed_byte_map = []
337
# a sorted list of keys matching each slice address for parsed bytes
338
# e.g. (None, 'foo@bar') would mean that the first byte contained no
339
# key, and the end byte of the slice is the of the data for 'foo@bar'
340
self._parsed_key_map = []
341
self._key_count = None
342
self._keys_by_offset = None
343
self._nodes_by_key = None
345
# The number of bytes we've read so far in trying to process this file
348
def __eq__(self, other):
349
"""Equal when self and other were created with the same parameters."""
351
type(self) == type(other) and
352
self._transport == other._transport and
353
self._name == other._name and
354
self._size == other._size)
356
def __ne__(self, other):
357
return not self.__eq__(other)
360
return "%s(%r)" % (self.__class__.__name__,
361
self._transport.abspath(self._name))
363
def _buffer_all(self, stream=None):
364
"""Buffer all the index data.
366
Mutates self._nodes and self.keys_by_offset.
368
if self._nodes is not None:
369
# We already did this
371
if 'index' in debug.debug_flags:
372
mutter('Reading entire index %s', self._transport.abspath(self._name))
374
stream = self._transport.get(self._name)
375
self._read_prefix(stream)
376
self._expected_elements = 3 + self._key_length
378
# raw data keyed by offset
379
self._keys_by_offset = {}
380
# ready-to-return key:value or key:value, node_ref_lists
382
self._nodes_by_key = None
385
lines = stream.read().split('\n')
387
_, _, _, trailers = self._parse_lines(lines, pos)
388
for key, absent, references, value in self._keys_by_offset.itervalues():
391
# resolve references:
392
if self.node_ref_lists:
393
node_value = (value, self._resolve_references(references))
396
self._nodes[key] = node_value
397
# cache the keys for quick set intersections
398
self._keys = set(self._nodes)
400
# there must be one line - the empty trailer line.
401
raise errors.BadIndexData(self)
403
def _get_nodes_by_key(self):
404
if self._nodes_by_key is None:
406
if self.node_ref_lists:
407
for key, (value, references) in self._nodes.iteritems():
408
key_dict = nodes_by_key
409
for subkey in key[:-1]:
410
key_dict = key_dict.setdefault(subkey, {})
411
key_dict[key[-1]] = key, value, references
413
for key, value in self._nodes.iteritems():
414
key_dict = nodes_by_key
415
for subkey in key[:-1]:
416
key_dict = key_dict.setdefault(subkey, {})
417
key_dict[key[-1]] = key, value
418
self._nodes_by_key = nodes_by_key
419
return self._nodes_by_key
421
def iter_all_entries(self):
422
"""Iterate over all keys within the index.
424
:return: An iterable of (index, key, value) or (index, key, value, reference_lists).
425
The former tuple is used when there are no reference lists in the
426
index, making the API compatible with simple key:value index types.
427
There is no defined order for the result iteration - it will be in
428
the most efficient order for the index.
430
if 'evil' in debug.debug_flags:
431
trace.mutter_callsite(3,
432
"iter_all_entries scales with size of history.")
433
if self._nodes is None:
435
if self.node_ref_lists:
436
for key, (value, node_ref_lists) in self._nodes.iteritems():
437
yield self, key, value, node_ref_lists
439
for key, value in self._nodes.iteritems():
440
yield self, key, value
442
def _read_prefix(self, stream):
443
signature = stream.read(len(self._signature()))
444
if not signature == self._signature():
445
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
446
options_line = stream.readline()
447
if not options_line.startswith(_OPTION_NODE_REFS):
448
raise errors.BadIndexOptions(self)
450
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):-1])
452
raise errors.BadIndexOptions(self)
453
options_line = stream.readline()
454
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
455
raise errors.BadIndexOptions(self)
457
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):-1])
459
raise errors.BadIndexOptions(self)
460
options_line = stream.readline()
461
if not options_line.startswith(_OPTION_LEN):
462
raise errors.BadIndexOptions(self)
464
self._key_count = int(options_line[len(_OPTION_LEN):-1])
466
raise errors.BadIndexOptions(self)
468
def _resolve_references(self, references):
469
"""Return the resolved key references for references.
471
References are resolved by looking up the location of the key in the
472
_keys_by_offset map and substituting the key name, preserving ordering.
474
:param references: An iterable of iterables of key locations. e.g.
476
:return: A tuple of tuples of keys.
479
for ref_list in references:
480
node_refs.append(tuple([self._keys_by_offset[ref][0] for ref in ref_list]))
481
return tuple(node_refs)
483
def _find_index(self, range_map, key):
484
"""Helper for the _parsed_*_index calls.
486
Given a range map - [(start, end), ...], finds the index of the range
487
in the map for key if it is in the map, and if it is not there, the
488
immediately preceeding range in the map.
490
result = bisect_right(range_map, key) - 1
491
if result + 1 < len(range_map):
492
# check the border condition, it may be in result + 1
493
if range_map[result + 1][0] == key[0]:
497
def _parsed_byte_index(self, offset):
498
"""Return the index of the entry immediately before offset.
500
e.g. if the parsed map has regions 0,10 and 11,12 parsed, meaning that
501
there is one unparsed byte (the 11th, addressed as[10]). then:
502
asking for 0 will return 0
503
asking for 10 will return 0
504
asking for 11 will return 1
505
asking for 12 will return 1
508
return self._find_index(self._parsed_byte_map, key)
510
def _parsed_key_index(self, key):
511
"""Return the index of the entry immediately before key.
513
e.g. if the parsed map has regions (None, 'a') and ('b','c') parsed,
514
meaning that keys from None to 'a' inclusive, and 'b' to 'c' inclusive
515
have been parsed, then:
516
asking for '' will return 0
517
asking for 'a' will return 0
518
asking for 'b' will return 1
519
asking for 'e' will return 1
521
search_key = (key, None)
522
return self._find_index(self._parsed_key_map, search_key)
524
def _is_parsed(self, offset):
525
"""Returns True if offset has been parsed."""
526
index = self._parsed_byte_index(offset)
527
if index == len(self._parsed_byte_map):
528
return offset < self._parsed_byte_map[index - 1][1]
529
start, end = self._parsed_byte_map[index]
530
return offset >= start and offset < end
532
def _iter_entries_from_total_buffer(self, keys):
533
"""Iterate over keys when the entire index is parsed."""
534
keys = keys.intersection(self._keys)
535
if self.node_ref_lists:
537
value, node_refs = self._nodes[key]
538
yield self, key, value, node_refs
541
yield self, key, self._nodes[key]
543
def iter_entries(self, keys):
544
"""Iterate over keys within the index.
546
:param keys: An iterable providing the keys to be retrieved.
547
:return: An iterable as per iter_all_entries, but restricted to the
548
keys supplied. No additional keys will be returned, and every
549
key supplied that is in the index will be returned.
554
if self._size is None and self._nodes is None:
557
# We fit about 20 keys per minimum-read (4K), so if we are looking for
558
# more than 1/20th of the index its likely (assuming homogenous key
559
# spread) that we'll read the entire index. If we're going to do that,
560
# buffer the whole thing. A better analysis might take key spread into
561
# account - but B+Tree indices are better anyway.
562
# We could look at all data read, and use a threshold there, which will
563
# trigger on ancestry walks, but that is not yet fully mapped out.
564
if self._nodes is None and len(keys) * 20 > self.key_count():
566
if self._nodes is not None:
567
return self._iter_entries_from_total_buffer(keys)
569
return (result[1] for result in bisect_multi_bytes(
570
self._lookup_keys_via_location, self._size, keys))
572
def iter_entries_prefix(self, keys):
573
"""Iterate over keys within the index using prefix matching.
575
Prefix matching is applied within the tuple of a key, not to within
576
the bytestring of each key element. e.g. if you have the keys ('foo',
577
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
578
only the former key is returned.
580
WARNING: Note that this method currently causes a full index parse
581
unconditionally (which is reasonably appropriate as it is a means for
582
thunking many small indices into one larger one and still supplies
583
iter_all_entries at the thunk layer).
585
:param keys: An iterable providing the key prefixes to be retrieved.
586
Each key prefix takes the form of a tuple the length of a key, but
587
with the last N elements 'None' rather than a regular bytestring.
588
The first element cannot be 'None'.
589
:return: An iterable as per iter_all_entries, but restricted to the
590
keys with a matching prefix to those supplied. No additional keys
591
will be returned, and every match that is in the index will be
597
# load data - also finds key lengths
598
if self._nodes is None:
600
if self._key_length == 1:
604
raise errors.BadIndexKey(key)
605
if len(key) != self._key_length:
606
raise errors.BadIndexKey(key)
607
if self.node_ref_lists:
608
value, node_refs = self._nodes[key]
609
yield self, key, value, node_refs
611
yield self, key, self._nodes[key]
613
nodes_by_key = self._get_nodes_by_key()
617
raise errors.BadIndexKey(key)
618
if len(key) != self._key_length:
619
raise errors.BadIndexKey(key)
620
# find what it refers to:
621
key_dict = nodes_by_key
623
# find the subdict whose contents should be returned.
625
while len(elements) and elements[0] is not None:
626
key_dict = key_dict[elements[0]]
629
# a non-existant lookup.
634
key_dict = dicts.pop(-1)
635
# can't be empty or would not exist
636
item, value = key_dict.iteritems().next()
637
if type(value) == dict:
639
dicts.extend(key_dict.itervalues())
642
for value in key_dict.itervalues():
643
# each value is the key:value:node refs tuple
645
yield (self, ) + value
647
# the last thing looked up was a terminal element
648
yield (self, ) + key_dict
651
"""Return an estimate of the number of keys in this index.
653
For GraphIndex the estimate is exact.
655
if self._key_count is None:
656
self._read_and_parse([_HEADER_READV])
657
return self._key_count
659
def _lookup_keys_via_location(self, location_keys):
660
"""Public interface for implementing bisection.
662
If _buffer_all has been called, then all the data for the index is in
663
memory, and this method should not be called, as it uses a separate
664
cache because it cannot pre-resolve all indices, which buffer_all does
667
:param location_keys: A list of location(byte offset), key tuples.
668
:return: A list of (location_key, result) tuples as expected by
669
bzrlib.bisect_multi.bisect_multi_bytes.
671
# Possible improvements:
672
# - only bisect lookup each key once
673
# - sort the keys first, and use that to reduce the bisection window
675
# this progresses in three parts:
678
# attempt to answer the question from the now in memory data.
679
# build the readv request
680
# for each location, ask for 800 bytes - much more than rows we've seen
683
for location, key in location_keys:
684
# can we answer from cache?
685
if self._bisect_nodes and key in self._bisect_nodes:
686
# We have the key parsed.
688
index = self._parsed_key_index(key)
689
if (len(self._parsed_key_map) and
690
self._parsed_key_map[index][0] <= key and
691
(self._parsed_key_map[index][1] >= key or
692
# end of the file has been parsed
693
self._parsed_byte_map[index][1] == self._size)):
694
# the key has been parsed, so no lookup is needed even if its
697
# - if we have examined this part of the file already - yes
698
index = self._parsed_byte_index(location)
699
if (len(self._parsed_byte_map) and
700
self._parsed_byte_map[index][0] <= location and
701
self._parsed_byte_map[index][1] > location):
702
# the byte region has been parsed, so no read is needed.
705
if location + length > self._size:
706
length = self._size - location
707
# todo, trim out parsed locations.
709
readv_ranges.append((location, length))
710
# read the header if needed
711
if self._bisect_nodes is None:
712
readv_ranges.append(_HEADER_READV)
713
self._read_and_parse(readv_ranges)
715
if self._nodes is not None:
716
# _read_and_parse triggered a _buffer_all because we requested the
718
for location, key in location_keys:
719
if key not in self._nodes: # not present
720
result.append(((location, key), False))
721
elif self.node_ref_lists:
722
value, refs = self._nodes[key]
723
result.append(((location, key),
724
(self, key, value, refs)))
726
result.append(((location, key),
727
(self, key, self._nodes[key])))
730
# - figure out <, >, missing, present
731
# - result present references so we can return them.
732
# keys that we cannot answer until we resolve references
733
pending_references = []
734
pending_locations = set()
735
for location, key in location_keys:
736
# can we answer from cache?
737
if key in self._bisect_nodes:
738
# the key has been parsed, so no lookup is needed
739
if self.node_ref_lists:
740
# the references may not have been all parsed.
741
value, refs = self._bisect_nodes[key]
742
wanted_locations = []
743
for ref_list in refs:
745
if ref not in self._keys_by_offset:
746
wanted_locations.append(ref)
748
pending_locations.update(wanted_locations)
749
pending_references.append((location, key))
751
result.append(((location, key), (self, key,
752
value, self._resolve_references(refs))))
754
result.append(((location, key),
755
(self, key, self._bisect_nodes[key])))
758
# has the region the key should be in, been parsed?
759
index = self._parsed_key_index(key)
760
if (self._parsed_key_map[index][0] <= key and
761
(self._parsed_key_map[index][1] >= key or
762
# end of the file has been parsed
763
self._parsed_byte_map[index][1] == self._size)):
764
result.append(((location, key), False))
766
# no, is the key above or below the probed location:
767
# get the range of the probed & parsed location
768
index = self._parsed_byte_index(location)
769
# if the key is below the start of the range, its below
770
if key < self._parsed_key_map[index][0]:
774
result.append(((location, key), direction))
776
# lookup data to resolve references
777
for location in pending_locations:
779
if location + length > self._size:
780
length = self._size - location
781
# TODO: trim out parsed locations (e.g. if the 800 is into the
782
# parsed region trim it, and dont use the adjust_for_latency
785
readv_ranges.append((location, length))
786
self._read_and_parse(readv_ranges)
787
if self._nodes is not None:
788
# The _read_and_parse triggered a _buffer_all, grab the data and
790
for location, key in pending_references:
791
value, refs = self._nodes[key]
792
result.append(((location, key), (self, key, value, refs)))
794
for location, key in pending_references:
795
# answer key references we had to look-up-late.
796
value, refs = self._bisect_nodes[key]
797
result.append(((location, key), (self, key,
798
value, self._resolve_references(refs))))
801
def _parse_header_from_bytes(self, bytes):
802
"""Parse the header from a region of bytes.
804
:param bytes: The data to parse.
805
:return: An offset, data tuple such as readv yields, for the unparsed
806
data. (which may length 0).
808
signature = bytes[0:len(self._signature())]
809
if not signature == self._signature():
810
raise errors.BadIndexFormatSignature(self._name, GraphIndex)
811
lines = bytes[len(self._signature()):].splitlines()
812
options_line = lines[0]
813
if not options_line.startswith(_OPTION_NODE_REFS):
814
raise errors.BadIndexOptions(self)
816
self.node_ref_lists = int(options_line[len(_OPTION_NODE_REFS):])
818
raise errors.BadIndexOptions(self)
819
options_line = lines[1]
820
if not options_line.startswith(_OPTION_KEY_ELEMENTS):
821
raise errors.BadIndexOptions(self)
823
self._key_length = int(options_line[len(_OPTION_KEY_ELEMENTS):])
825
raise errors.BadIndexOptions(self)
826
options_line = lines[2]
827
if not options_line.startswith(_OPTION_LEN):
828
raise errors.BadIndexOptions(self)
830
self._key_count = int(options_line[len(_OPTION_LEN):])
832
raise errors.BadIndexOptions(self)
833
# calculate the bytes we have processed
834
header_end = (len(signature) + len(lines[0]) + len(lines[1]) +
836
self._parsed_bytes(0, None, header_end, None)
837
# setup parsing state
838
self._expected_elements = 3 + self._key_length
839
# raw data keyed by offset
840
self._keys_by_offset = {}
841
# keys with the value and node references
842
self._bisect_nodes = {}
843
return header_end, bytes[header_end:]
845
def _parse_region(self, offset, data):
846
"""Parse node data returned from a readv operation.
848
:param offset: The byte offset the data starts at.
849
:param data: The data to parse.
853
end = offset + len(data)
856
# Trivial test - if the current index's end is within the
857
# low-matching parsed range, we're done.
858
index = self._parsed_byte_index(high_parsed)
859
if end < self._parsed_byte_map[index][1]:
861
# print "[%d:%d]" % (offset, end), \
862
# self._parsed_byte_map[index:index + 2]
863
high_parsed, last_segment = self._parse_segment(
864
offset, data, end, index)
868
def _parse_segment(self, offset, data, end, index):
869
"""Parse one segment of data.
871
:param offset: Where 'data' begins in the file.
872
:param data: Some data to parse a segment of.
873
:param end: Where data ends
874
:param index: The current index into the parsed bytes map.
875
:return: True if the parsed segment is the last possible one in the
877
:return: high_parsed_byte, last_segment.
878
high_parsed_byte is the location of the highest parsed byte in this
879
segment, last_segment is True if the parsed segment is the last
880
possible one in the data block.
882
# default is to use all data
884
# accomodate overlap with data before this.
885
if offset < self._parsed_byte_map[index][1]:
886
# overlaps the lower parsed region
887
# skip the parsed data
888
trim_start = self._parsed_byte_map[index][1] - offset
889
# don't trim the start for \n
890
start_adjacent = True
891
elif offset == self._parsed_byte_map[index][1]:
892
# abuts the lower parsed region
895
# do not trim anything
896
start_adjacent = True
898
# does not overlap the lower parsed region
901
# but trim the leading \n
902
start_adjacent = False
903
if end == self._size:
904
# lines up to the end of all data:
907
# do not strip to the last \n
910
elif index + 1 == len(self._parsed_byte_map):
911
# at the end of the parsed data
914
# but strip to the last \n
917
elif end == self._parsed_byte_map[index + 1][0]:
918
# buts up against the next parsed region
921
# do not strip to the last \n
924
elif end > self._parsed_byte_map[index + 1][0]:
925
# overlaps into the next parsed region
926
# only consider the unparsed data
927
trim_end = self._parsed_byte_map[index + 1][0] - offset
928
# do not strip to the last \n as we know its an entire record
930
last_segment = end < self._parsed_byte_map[index + 1][1]
932
# does not overlap into the next region
935
# but strip to the last \n
938
# now find bytes to discard if needed
939
if not start_adjacent:
940
# work around python bug in rfind
941
if trim_start is None:
942
trim_start = data.find('\n') + 1
944
trim_start = data.find('\n', trim_start) + 1
945
if not (trim_start != 0):
946
raise AssertionError('no \n was present')
947
# print 'removing start', offset, trim_start, repr(data[:trim_start])
949
# work around python bug in rfind
951
trim_end = data.rfind('\n') + 1
953
trim_end = data.rfind('\n', None, trim_end) + 1
954
if not (trim_end != 0):
955
raise AssertionError('no \n was present')
956
# print 'removing end', offset, trim_end, repr(data[trim_end:])
957
# adjust offset and data to the parseable data.
958
trimmed_data = data[trim_start:trim_end]
959
if not (trimmed_data):
960
raise AssertionError('read unneeded data [%d:%d] from [%d:%d]'
961
% (trim_start, trim_end, offset, offset + len(data)))
964
# print "parsing", repr(trimmed_data)
965
# splitlines mangles the \r delimiters.. don't use it.
966
lines = trimmed_data.split('\n')
969
first_key, last_key, nodes, _ = self._parse_lines(lines, pos)
970
for key, value in nodes:
971
self._bisect_nodes[key] = value
972
self._parsed_bytes(offset, first_key,
973
offset + len(trimmed_data), last_key)
974
return offset + len(trimmed_data), last_segment
976
def _parse_lines(self, lines, pos):
985
if not (self._size == pos + 1):
986
raise AssertionError("%s %s" % (self._size, pos))
989
elements = line.split('\0')
990
if len(elements) != self._expected_elements:
991
raise errors.BadIndexData(self)
992
# keys are tuples. Each element is a string that may occur many
993
# times, so we intern them to save space. AB, RC, 200807
994
key = tuple([intern(element) for element in elements[:self._key_length]])
995
if first_key is None:
997
absent, references, value = elements[-3:]
999
for ref_string in references.split('\t'):
1000
ref_lists.append(tuple([
1001
int(ref) for ref in ref_string.split('\r') if ref
1003
ref_lists = tuple(ref_lists)
1004
self._keys_by_offset[pos] = (key, absent, ref_lists, value)
1005
pos += len(line) + 1 # +1 for the \n
1008
if self.node_ref_lists:
1009
node_value = (value, ref_lists)
1012
nodes.append((key, node_value))
1013
# print "parsed ", key
1014
return first_key, key, nodes, trailers
1016
def _parsed_bytes(self, start, start_key, end, end_key):
1017
"""Mark the bytes from start to end as parsed.
1019
Calling self._parsed_bytes(1,2) will mark one byte (the one at offset
1022
:param start: The start of the parsed region.
1023
:param end: The end of the parsed region.
1025
index = self._parsed_byte_index(start)
1026
new_value = (start, end)
1027
new_key = (start_key, end_key)
1029
# first range parsed is always the beginning.
1030
self._parsed_byte_map.insert(index, new_value)
1031
self._parsed_key_map.insert(index, new_key)
1035
# extend lower region
1036
# extend higher region
1037
# combine two regions
1038
if (index + 1 < len(self._parsed_byte_map) and
1039
self._parsed_byte_map[index][1] == start and
1040
self._parsed_byte_map[index + 1][0] == end):
1041
# combine two regions
1042
self._parsed_byte_map[index] = (self._parsed_byte_map[index][0],
1043
self._parsed_byte_map[index + 1][1])
1044
self._parsed_key_map[index] = (self._parsed_key_map[index][0],
1045
self._parsed_key_map[index + 1][1])
1046
del self._parsed_byte_map[index + 1]
1047
del self._parsed_key_map[index + 1]
1048
elif self._parsed_byte_map[index][1] == start:
1049
# extend the lower entry
1050
self._parsed_byte_map[index] = (
1051
self._parsed_byte_map[index][0], end)
1052
self._parsed_key_map[index] = (
1053
self._parsed_key_map[index][0], end_key)
1054
elif (index + 1 < len(self._parsed_byte_map) and
1055
self._parsed_byte_map[index + 1][0] == end):
1056
# extend the higher entry
1057
self._parsed_byte_map[index + 1] = (
1058
start, self._parsed_byte_map[index + 1][1])
1059
self._parsed_key_map[index + 1] = (
1060
start_key, self._parsed_key_map[index + 1][1])
1063
self._parsed_byte_map.insert(index + 1, new_value)
1064
self._parsed_key_map.insert(index + 1, new_key)
1066
def _read_and_parse(self, readv_ranges):
1067
"""Read the the ranges and parse the resulting data.
1069
:param readv_ranges: A prepared readv range list.
1071
if not readv_ranges:
1073
if self._nodes is None and self._bytes_read * 2 >= self._size:
1074
# We've already read more than 50% of the file and we are about to
1075
# request more data, just _buffer_all() and be done
1079
readv_data = self._transport.readv(self._name, readv_ranges, True,
1082
for offset, data in readv_data:
1083
self._bytes_read += len(data)
1084
if offset == 0 and len(data) == self._size:
1085
# We read the whole range, most likely because the
1086
# Transport upcast our readv ranges into one long request
1087
# for enough total data to grab the whole index.
1088
self._buffer_all(StringIO(data))
1090
if self._bisect_nodes is None:
1091
# this must be the start
1092
if not (offset == 0):
1093
raise AssertionError()
1094
offset, data = self._parse_header_from_bytes(data)
1095
# print readv_ranges, "[%d:%d]" % (offset, offset + len(data))
1096
self._parse_region(offset, data)
1098
def _signature(self):
1099
"""The file signature for this index type."""
1103
"""Validate that everything in the index can be accessed."""
1104
# iter_all validates completely at the moment, so just do that.
1105
for node in self.iter_all_entries():
1109
class CombinedGraphIndex(object):
1110
"""A GraphIndex made up from smaller GraphIndices.
1112
The backing indices must implement GraphIndex, and are presumed to be
1115
Queries against the combined index will be made against the first index,
1116
and then the second and so on. The order of index's can thus influence
1117
performance significantly. For example, if one index is on local disk and a
1118
second on a remote server, the local disk index should be before the other
1122
def __init__(self, indices, reload_func=None):
1123
"""Create a CombinedGraphIndex backed by indices.
1125
:param indices: An ordered list of indices to query for data.
1126
:param reload_func: A function to call if we find we are missing an
1127
index. Should have the form reload_func() => True/False to indicate
1128
if reloading actually changed anything.
1130
self._indices = indices
1131
self._reload_func = reload_func
1135
self.__class__.__name__,
1136
', '.join(map(repr, self._indices)))
1138
@symbol_versioning.deprecated_method(symbol_versioning.one_one)
1139
def get_parents(self, revision_ids):
1140
"""See graph._StackedParentsProvider.get_parents.
1142
This implementation thunks the graph.Graph.get_parents api across to
1145
:param revision_ids: An iterable of graph keys for this graph.
1146
:return: A list of parent details for each key in revision_ids.
1147
Each parent details will be one of:
1148
* None when the key was missing
1149
* (NULL_REVISION,) when the key has no parents.
1150
* (parent_key, parent_key...) otherwise.
1152
parent_map = self.get_parent_map(revision_ids)
1153
return [parent_map.get(r, None) for r in revision_ids]
1155
def get_parent_map(self, keys):
1156
"""See graph._StackedParentsProvider.get_parent_map"""
1157
search_keys = set(keys)
1158
if NULL_REVISION in search_keys:
1159
search_keys.discard(NULL_REVISION)
1160
found_parents = {NULL_REVISION:[]}
1163
for index, key, value, refs in self.iter_entries(search_keys):
1166
parents = (NULL_REVISION,)
1167
found_parents[key] = parents
1168
return found_parents
1170
def insert_index(self, pos, index):
1171
"""Insert a new index in the list of indices to query.
1173
:param pos: The position to insert the index.
1174
:param index: The index to insert.
1176
self._indices.insert(pos, index)
1178
def iter_all_entries(self):
1179
"""Iterate over all keys within the index
1181
Duplicate keys across child indices are presumed to have the same
1182
value and are only reported once.
1184
:return: An iterable of (index, key, reference_lists, value).
1185
There is no defined order for the result iteration - it will be in
1186
the most efficient order for the index.
1191
for index in self._indices:
1192
for node in index.iter_all_entries():
1193
if node[1] not in seen_keys:
1195
seen_keys.add(node[1])
1197
except errors.NoSuchFile:
1198
self._reload_or_raise()
1200
def iter_entries(self, keys):
1201
"""Iterate over keys within the index.
1203
Duplicate keys across child indices are presumed to have the same
1204
value and are only reported once.
1206
:param keys: An iterable providing the keys to be retrieved.
1207
:return: An iterable of (index, key, reference_lists, value). There is no
1208
defined order for the result iteration - it will be in the most
1209
efficient order for the index.
1214
for index in self._indices:
1217
for node in index.iter_entries(keys):
1218
keys.remove(node[1])
1221
except errors.NoSuchFile:
1222
self._reload_or_raise()
1224
def iter_entries_prefix(self, keys):
1225
"""Iterate over keys within the index using prefix matching.
1227
Duplicate keys across child indices are presumed to have the same
1228
value and are only reported once.
1230
Prefix matching is applied within the tuple of a key, not to within
1231
the bytestring of each key element. e.g. if you have the keys ('foo',
1232
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1233
only the former key is returned.
1235
:param keys: An iterable providing the key prefixes to be retrieved.
1236
Each key prefix takes the form of a tuple the length of a key, but
1237
with the last N elements 'None' rather than a regular bytestring.
1238
The first element cannot be 'None'.
1239
:return: An iterable as per iter_all_entries, but restricted to the
1240
keys with a matching prefix to those supplied. No additional keys
1241
will be returned, and every match that is in the index will be
1250
for index in self._indices:
1251
for node in index.iter_entries_prefix(keys):
1252
if node[1] in seen_keys:
1254
seen_keys.add(node[1])
1257
except errors.NoSuchFile:
1258
self._reload_or_raise()
1260
def key_count(self):
1261
"""Return an estimate of the number of keys in this index.
1263
For CombinedGraphIndex this is approximated by the sum of the keys of
1264
the child indices. As child indices may have duplicate keys this can
1265
have a maximum error of the number of child indices * largest number of
1270
return sum((index.key_count() for index in self._indices), 0)
1271
except errors.NoSuchFile:
1272
self._reload_or_raise()
1274
def _reload_or_raise(self):
1275
"""We just got a NoSuchFile exception.
1277
Try to reload the indices, if it fails, just raise the current
1280
if self._reload_func is None:
1282
exc_type, exc_value, exc_traceback = sys.exc_info()
1283
trace.mutter('Trying to reload after getting exception: %s',
1285
if not self._reload_func():
1286
# We tried to reload, but nothing changed, so we fail anyway
1287
trace.mutter('_reload_func indicated nothing has changed.'
1288
' Raising original exception.')
1289
raise exc_type, exc_value, exc_traceback
1292
"""Validate that everything in the index can be accessed."""
1295
for index in self._indices:
1298
except errors.NoSuchFile:
1299
self._reload_or_raise()
1302
class InMemoryGraphIndex(GraphIndexBuilder):
1303
"""A GraphIndex which operates entirely out of memory and is mutable.
1305
This is designed to allow the accumulation of GraphIndex entries during a
1306
single write operation, where the accumulated entries need to be immediately
1307
available - for example via a CombinedGraphIndex.
1310
def add_nodes(self, nodes):
1311
"""Add nodes to the index.
1313
:param nodes: An iterable of (key, node_refs, value) entries to add.
1315
if self.reference_lists:
1316
for (key, value, node_refs) in nodes:
1317
self.add_node(key, value, node_refs)
1319
for (key, value) in nodes:
1320
self.add_node(key, value)
1322
def iter_all_entries(self):
1323
"""Iterate over all keys within the index
1325
:return: An iterable of (index, key, reference_lists, value). There is no
1326
defined order for the result iteration - it will be in the most
1327
efficient order for the index (in this case dictionary hash order).
1329
if 'evil' in debug.debug_flags:
1330
trace.mutter_callsite(3,
1331
"iter_all_entries scales with size of history.")
1332
if self.reference_lists:
1333
for key, (absent, references, value) in self._nodes.iteritems():
1335
yield self, key, value, references
1337
for key, (absent, references, value) in self._nodes.iteritems():
1339
yield self, key, value
1341
def iter_entries(self, keys):
1342
"""Iterate over keys within the index.
1344
:param keys: An iterable providing the keys to be retrieved.
1345
:return: An iterable of (index, key, value, reference_lists). There is no
1346
defined order for the result iteration - it will be in the most
1347
efficient order for the index (keys iteration order in this case).
1350
if self.reference_lists:
1351
for key in keys.intersection(self._keys):
1352
node = self._nodes[key]
1354
yield self, key, node[2], node[1]
1356
for key in keys.intersection(self._keys):
1357
node = self._nodes[key]
1359
yield self, key, node[2]
1361
def iter_entries_prefix(self, keys):
1362
"""Iterate over keys within the index using prefix matching.
1364
Prefix matching is applied within the tuple of a key, not to within
1365
the bytestring of each key element. e.g. if you have the keys ('foo',
1366
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1367
only the former key is returned.
1369
:param keys: An iterable providing the key prefixes to be retrieved.
1370
Each key prefix takes the form of a tuple the length of a key, but
1371
with the last N elements 'None' rather than a regular bytestring.
1372
The first element cannot be 'None'.
1373
:return: An iterable as per iter_all_entries, but restricted to the
1374
keys with a matching prefix to those supplied. No additional keys
1375
will be returned, and every match that is in the index will be
1378
# XXX: To much duplication with the GraphIndex class; consider finding
1379
# a good place to pull out the actual common logic.
1383
if self._key_length == 1:
1387
raise errors.BadIndexKey(key)
1388
if len(key) != self._key_length:
1389
raise errors.BadIndexKey(key)
1390
node = self._nodes[key]
1393
if self.reference_lists:
1394
yield self, key, node[2], node[1]
1396
yield self, key, node[2]
1398
nodes_by_key = self._get_nodes_by_key()
1402
raise errors.BadIndexKey(key)
1403
if len(key) != self._key_length:
1404
raise errors.BadIndexKey(key)
1405
# find what it refers to:
1406
key_dict = nodes_by_key
1407
elements = list(key)
1408
# find the subdict to return
1410
while len(elements) and elements[0] is not None:
1411
key_dict = key_dict[elements[0]]
1414
# a non-existant lookup.
1419
key_dict = dicts.pop(-1)
1420
# can't be empty or would not exist
1421
item, value = key_dict.iteritems().next()
1422
if type(value) == dict:
1424
dicts.extend(key_dict.itervalues())
1427
for value in key_dict.itervalues():
1428
yield (self, ) + value
1430
yield (self, ) + key_dict
1432
def key_count(self):
1433
"""Return an estimate of the number of keys in this index.
1435
For InMemoryGraphIndex the estimate is exact.
1437
return len(self._keys)
1440
"""In memory index's have no known corruption at the moment."""
1443
class GraphIndexPrefixAdapter(object):
1444
"""An adapter between GraphIndex with different key lengths.
1446
Queries against this will emit queries against the adapted Graph with the
1447
prefix added, queries for all items use iter_entries_prefix. The returned
1448
nodes will have their keys and node references adjusted to remove the
1449
prefix. Finally, an add_nodes_callback can be supplied - when called the
1450
nodes and references being added will have prefix prepended.
1453
def __init__(self, adapted, prefix, missing_key_length,
1454
add_nodes_callback=None):
1455
"""Construct an adapter against adapted with prefix."""
1456
self.adapted = adapted
1457
self.prefix_key = prefix + (None,)*missing_key_length
1458
self.prefix = prefix
1459
self.prefix_len = len(prefix)
1460
self.add_nodes_callback = add_nodes_callback
1462
def add_nodes(self, nodes):
1463
"""Add nodes to the index.
1465
:param nodes: An iterable of (key, node_refs, value) entries to add.
1467
# save nodes in case its an iterator
1468
nodes = tuple(nodes)
1469
translated_nodes = []
1471
# Add prefix_key to each reference node_refs is a tuple of tuples,
1472
# so split it apart, and add prefix_key to the internal reference
1473
for (key, value, node_refs) in nodes:
1474
adjusted_references = (
1475
tuple(tuple(self.prefix + ref_node for ref_node in ref_list)
1476
for ref_list in node_refs))
1477
translated_nodes.append((self.prefix + key, value,
1478
adjusted_references))
1480
# XXX: TODO add an explicit interface for getting the reference list
1481
# status, to handle this bit of user-friendliness in the API more
1483
for (key, value) in nodes:
1484
translated_nodes.append((self.prefix + key, value))
1485
self.add_nodes_callback(translated_nodes)
1487
def add_node(self, key, value, references=()):
1488
"""Add a node to the index.
1490
:param key: The key. keys are non-empty tuples containing
1491
as many whitespace-free utf8 bytestrings as the key length
1492
defined for this index.
1493
:param references: An iterable of iterables of keys. Each is a
1494
reference to another key.
1495
:param value: The value to associate with the key. It may be any
1496
bytes as long as it does not contain \0 or \n.
1498
self.add_nodes(((key, value, references), ))
1500
def _strip_prefix(self, an_iter):
1501
"""Strip prefix data from nodes and return it."""
1502
for node in an_iter:
1504
if node[1][:self.prefix_len] != self.prefix:
1505
raise errors.BadIndexData(self)
1506
for ref_list in node[3]:
1507
for ref_node in ref_list:
1508
if ref_node[:self.prefix_len] != self.prefix:
1509
raise errors.BadIndexData(self)
1510
yield node[0], node[1][self.prefix_len:], node[2], (
1511
tuple(tuple(ref_node[self.prefix_len:] for ref_node in ref_list)
1512
for ref_list in node[3]))
1514
def iter_all_entries(self):
1515
"""Iterate over all keys within the index
1517
iter_all_entries is implemented against the adapted index using
1518
iter_entries_prefix.
1520
:return: An iterable of (index, key, reference_lists, value). There is no
1521
defined order for the result iteration - it will be in the most
1522
efficient order for the index (in this case dictionary hash order).
1524
return self._strip_prefix(self.adapted.iter_entries_prefix([self.prefix_key]))
1526
def iter_entries(self, keys):
1527
"""Iterate over keys within the index.
1529
:param keys: An iterable providing the keys to be retrieved.
1530
:return: An iterable of (index, key, value, reference_lists). There is no
1531
defined order for the result iteration - it will be in the most
1532
efficient order for the index (keys iteration order in this case).
1534
return self._strip_prefix(self.adapted.iter_entries(
1535
self.prefix + key for key in keys))
1537
def iter_entries_prefix(self, keys):
1538
"""Iterate over keys within the index using prefix matching.
1540
Prefix matching is applied within the tuple of a key, not to within
1541
the bytestring of each key element. e.g. if you have the keys ('foo',
1542
'bar'), ('foobar', 'gam') and do a prefix search for ('foo', None) then
1543
only the former key is returned.
1545
:param keys: An iterable providing the key prefixes to be retrieved.
1546
Each key prefix takes the form of a tuple the length of a key, but
1547
with the last N elements 'None' rather than a regular bytestring.
1548
The first element cannot be 'None'.
1549
:return: An iterable as per iter_all_entries, but restricted to the
1550
keys with a matching prefix to those supplied. No additional keys
1551
will be returned, and every match that is in the index will be
1554
return self._strip_prefix(self.adapted.iter_entries_prefix(
1555
self.prefix + key for key in keys))
1557
def key_count(self):
1558
"""Return an estimate of the number of keys in this index.
1560
For GraphIndexPrefixAdapter this is relatively expensive - key
1561
iteration with the prefix is done.
1563
return len(list(self.iter_all_entries()))
1566
"""Call the adapted's validate."""
1567
self.adapted.validate()
added
removed
bzrlib/index.py
